Image Preprocessing for OCR Text Alignment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Legacy and contemporary OCR software face difficulties with non-vertically aligned text, leading to erroneous OCR results, particularly in documents like forms and spreadsheets, due to the raster scanning approach which fails to consistently bisect text at uniform heights, resulting in mis-read, double-read, and missing text issues.
Innovation Solution
A processor-based system that pre-processes images to vertically align row-based text by identifying and adjusting the position of text pixels within calculated bounding boxes, ensuring consistent vertical alignment before OCR processing, thereby improving the accuracy and quality of OCR output without requiring updates to existing OCR software.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional raster scanning approach is used in OCR software, then the software can process text in a straightforward manner, but it fails to handle non-vertically aligned text correctly, leading to erroneous OCR results
Solution Approach 1:
The patent applies preliminary action by performing vertical alignment of text rows before the OCR recognition process. The system calculates the vertical position of each text row and adjusts their positions to ensure uniform alignment along the vertical axis. This preprocessing step ensures that when the raster scanner subsequently processes the image, all text in a row is properly aligned, eliminating the errors that would otherwise occur with non-vertically aligned text.
2Reliability
If sophisticated algorithms are designed to accommodate non-vertically aligned text within OCR software, then OCR accuracy for such text would improve, but the device complexity and software update requirements increase
Solution Approach 1:
The patent extracts the vertical alignment function from the core OCR recognition process and implements it as a separate preprocessing step. Instead of modifying the complex OCR engine to handle non-aligned text, the system isolates the alignment problem and solves it independently before feeding the aligned image to the existing OCR software. This maintains the simplicity of the OCR core while adding the necessary capability to handle non-vertically aligned text.
3Productivity
If the raster scanner bisects text at varying heights due to non-vertical alignment, then processing speed is maintained, but text recognition accuracy deteriorates with mis-read, double-read, and missing text issues
Solution Approach 1:
The system performs vertical alignment as a preliminary action before raster scanning, ensuring that all text rows are properly positioned. This allows the raster scanner to maintain its high processing speed while accurately capturing text at consistent heights, eliminating mis-reads, double-reads, and missing text issues that would otherwise occur.
Data Source
AI summary
A system to preprocess images to increase accuracy of optical character recognition (OCR) includes a processor, and a memory coupled to the processor. The processor is configured to scan an electronically stored representation of a whole or partial document, identify an image in the electronically stored representation, and recognize row-based text within the electronically stored representation. In addition, the processor is configured to align the row-based text vertically, generate a resultant electronically stored representation of the whole or partial document having the row-based text aligned, and save the resultant electronically stored representation for subsequent OCR processing. The electronically stored representation of the whole or partial document may contain at least one image having a JPG, TIF, GIF, PNG, or BMP, type of format.


